Professional Documents
Culture Documents
computing approach
Yahui Chai,1, ∗ Arianna Crippa,1, 2 Karl Jansen,1, 3 Stefan Kühn,1
Vincent R. Pascuzzi,4, † Francesco Tacchino,5 and Ivano Tavernelli5
1
CQTA, Deutsches Elektronen-Synchrotron DESY, Platanenallee 6, 15738 Zeuthen, Germany
2
Institut für Physik, Humboldt-Universität zu Berlin, Newtonstr. 15, 12489 Berlin, Germany
3
Computation-Based Science and Technology Research Center,
The Cyprus Institute, 20 Kavafi Street, 2121 Nicosia, Cyprus
4
IBM Quantum, IBM Thomas J. Watson Research Center, Yorktown Heights, NY 10598, USA
5
IBM Quantum, IBM Research Europe – Zurich, Rueschlikon 8803, Switzerland
(Dated: December 6, 2023)
We propose a method to prepare Gaussian wave packets with momentum on top of the interacting
arXiv:2312.02272v1 [quant-ph] 4 Dec 2023
ground state of a fermionic Hamiltonian. Using Givens rotation, we show how to efficiently obtain
expectation values of observables throughout the evolution of the wave packets on digital quantum
computers. We demonstrate our technique by applying it to the staggered lattice formulation of
the Thirring model and studying the scattering of two wave packets. Monitoring the the particle
density and the entropy produced during the scattering process, we characterize the phenomenon
and provide a first step towards studying more complicated collision processes on digital quantum
computers. In addition, we perform a small-scale demonstration on IBM’s quantum hardware,
showing that our method is suitable for current and near-term quantum devices.
i
where we again identify σN = σ0i , i ∈ {x, y, z}, and ten- non-interacting case, they will also provide a valid ap-
sor products between operators are assumed. Note that proximation for a particle in the interacting case. In Sec-
the non-local Pauli-strings in the second line of Eq. (5) tion III B we then discuss the implementation of these
arise due to the periodic boundary conditions, and are a states on a quantum device and provide the correspond-
consequence
of the Jordan-Wigner
transformation of the ing circuits using Givens rotations entangling gates. The
† †
terms ξ0 ξN −1 − ξN −1 ξ0 wrapping around the bound- implementation of the quantum dynamics circuits is de-
scribed in Section III C, while Section III D reports on
ary. For the rest of this paper, we consider without loss
the calculation of the time-dependent observables.
of generality a = 1.
FIG. 1. Probability density for a Gaussian wave packet in momentum and position space for N = 50 sites. The upper row shows
a Gaussian fermionic wave packet with µck = 5×2π/N , the lower row a Gaussian antifermion wave packet with µdk = −5×2π/N ,
both with a width of σk = 2π/N . Panels (a) and (d) correspond to the distribution in momentum space, panels (b), (c) and (e),
(f) show the probability density of the single Gaussian (anti)fermion in position space for masses m = 1 and 100. The second
and third column illustrate that with increasing mass, resulting in a value of vk closer to zero, the probabiltiy distribution for
the fermion (antifermion) in position space is increasingly suppressed on odd (even) sites.
where ϕd∗ d
n is the complex conjugate of ϕn , and we trans-
† θn x y y x
V (rn (θn )) = exp i σn−1 σn − σn−1 σn . (21) lated the fermionic operators from Eq. (15) to spin op-
2
erators, again using the Jordan-Wigner transformation.
The two-qubit gate in Eq. (21) above can be easily ex- The unitary operators V on the right-hand side can be
pressed by single-qubit Pauli rotation gates, RP (θ) = implemented as quantum circuit with a combination of
exp(−iP θ/2), P ∈ {σ x , σ y , σ z }, and CNOT gates as Givens rotations [27–29], as outlined above. Note that
shown in Fig. 2. Analogously, we can obtain an expres- the Pauli raising operator σ0+ and its Hermitian conju-
6
gate are non-unitary and therefore cannot be directly im- Eqs. (22) and (23),
plemented on a quantum device. Nevertheless, the form
of the operators C † (ϕc ) and D† (ϕd ) we derived here is |ψ(t)⟩ = U3 σ0+ U2 σ0− U1 |0⟩
advantageous to obtain the time-dependent expectation 1 (24)
values observables on a quantum device, as discussed in = U3 (σ0x + iσ0y )U2 (σ0x − iσ0y )U1 |0⟩ ,
4
Sec. III D.
where we have defined the unitaries
U3 = e−iHt V (ϕd∗
n ), (25)
†
C. State initialization and simulation of dynamics U2 = V (ϕd∗ c
n )V (ϕn ), (26)
† c
U1 = V (ϕn )U0 , (27)
In order to simulate the dynamics of scattering pro-
cesses, the initial state we want to generate is given by and used the fact that the ground state can be prepared
|ψ(0)⟩ = D† (ϕdn )C † (ϕcn ) |Ω⟩. For the free fermionic case via some unitary U0 from the state |0⟩. Note that the
discussed in Sec. III A, which corresponds to the Thirring state |ψ(t)⟩ in Eq. (24) is not a normalized state that can
model with coupling g = 0, this state exactly represents be prepared on a quantum device, as the operators σ0± =
the ground state with two Gaussian wave packets – one (σ0x ± iσ0y )/2 are not unitary. Nevertheless, it is possible
corresponding to an antifermion and another correspond- to obtain expected values of observables efficiently from
ing to a fermion – superimposed on top. For the case of a quantum device, as we discuss in the following section.
non-vanishing coupling, applying the operators D† (ϕdn )
and C † (ϕcn ) to the ground state will no longer produce ex-
D. Evaluation of time-dependent observables
actly a state with a particle-antiparticle wave packet, but
should provide a good approximation. This will be veri-
fied a posteriori in our numerical simulations in Sec. IV. In general, we want to calculate the expectation value
of some observable O in the state |ψ(t)⟩. Using Eq. (24),
Here we focus on the description of the dynamics of
the expectation value of O can be expressed as
such an initial state, which is given by
1 1 1 X
−iHt −iHt † ⟨O⟩ = ⟨ψ(t)| O |ψ(t)⟩ = × cµ̄ ⟨O⟩µ̄ , (28)
|ψ(t)⟩ = e |ψ(0)⟩ = e D (ϕdn )C † (ϕcn ) |Ω⟩ . (23) N N 16 µ̄
As mentioned before, the dynamics can be implemented where we have explicitly taken into account the normal-
via standard algorithms, such as Trotter decomposition ization factor N = ⟨ψ(t)|ψ(t)⟩ since |ψ(t)⟩ is not nor-
or variational time evolution. We will not discuss these malized. The symbol µ̄ corresponds to a multi-index
approaches in detail, and estimating the resources re- µ̄ = (µ1 , µ2 , µ3 , µ4 ) with µi ∈ {x, y}, and ⟨O⟩µ̄ is a short-
quired for these methods as well as assessing their per- hand notation for
formance for the general case is left for future work.
⟨O⟩µ̄ = ⟨0| (U1† σ0µ1 U2† σ0µ2 U3† ) O (U3 σ0µ3 U2 σ0µ4 U1 ) |0⟩ .
For Hamiltonians that are at most quadratic in the (29)
the fermionic operators, the time evolution can be im-
plemented more efficiently. On the one hand, the con- The coefficients cµ̄ in Eq. (28) are given by {±1, ±i}, de-
sidered Hamiltonian is diagonal in momentum space. pending on the combination of Pauli matrices involved.
Hence, if the evolution is computed in momentum space, Note that in Eq. (29) all operators acting on the state
exp(−iHt) simply corresponds to a phase factor. On the |0⟩ are unitary and can be implemented a quantum com-
other hand, in position space, exp(−iHt) has the form of puter.
Eq. (16). Thus, it can be decomposed using Givens rota- For the case µ1 = µ4 and µ2 = µ3 , Eq. (28) repre-
tions as demonstrated for the operators creating wave sents nothing but the expectation value of observable O
packets in the previous section. The resulting circuit in the state |χ⟩ = U3 σ0µ2 U2 σ0µ1 U1 |0⟩, which can be di-
implements the time evolution operator for arbitrary t rectly measured on a quantum device.
without any approximation at a constant circuit depth In all other cases, we can find the expected value of the
O(4N ) using O(N 2 ) CNOT gates. Thus, for the non- non-Hermitian operator with the following observation.
interacting case this approach presents a more promis- Since we sum over all possible combinations in Eq. (29)
ing avenue towards simulating the dynamics on quantum and the Pauli matrices are self-adjoint, the sum is going
hardware than using methods based on Trotter decompo- to contain the Hermitian conjugate of each summand.
sition or variational time evolution. The technical details Hence, we do not need to determine the individual sum-
for implementing the time evolution operator in the non- mands; it is sufficient to obtain (cµ̄ ⟨O⟩µ̄ + h.c.) from the
interacting case using Givens rotation are presented in quantum device. This can be done with a circuit sim-
Appendix C. ilar to that used for the Hadamard test [47, 48]. The
In general, we find the time-evolved state by combining corresponding quantum circuit requires the addition of
7
where S1,(C,D) (n, t) is the entropy for the time evolved progress faster for smaller values of the mass, as expected.
state ψ(C,D) (0) , according to Eq. (32). Since the fermions are noninteracting, the wave packets
simply pass through each other before approaching one
boundary and returning on the opposite due to periodic
A. Vanishing coupling: the free fermionic case boundary conditions.
In order to gain more insight into the scattering pro-
Let us first consider the case of free staggered Dirac cess, we also consider the von Neumann entropy. Since
fermions. While the dynamics of the system in this computing the entropy in our direct approach requires
regime can be simulated efficiently in momentum space constructing the reduced density operator, which in gen-
using free fermionic techniques (see Appendix B), we eral is a dense matrix, we use a slightly smaller system
choose to work in position space to benchmark the tech- size N = 14 for computational convenience. Moreover,
niques developed earlier. we do not perform a direct simulation of the state vector
In order to demonstrate that the operators in Eq. (22) as in the previously, but we simulate the the circuit al-
successfully create a wave packet, we first simulate the lowing us to compute the evolution exactly as discussed
evolution in time exactly. To this end, we obtain the in Sec. III C and outlined in detail in Appendix C. For
ground state |Ω⟩ of the Hamiltonian using exact diago- these simulations, we use Qiskit [58] assuming an ideal
nalization and subsequently compute the system’s evo- quantum computer without shot noise. The results for
lution via Eq. (23) by approximating the time evolu- the subtracted particle density and the von Neumann
tion operator via a Taylor expansion up to second order, entropy are shown in Fig. 5. Despite the reduced sys-
exp(−iH∆t) ≈ 1 − i∆tH − H 2 (∆t)2 /2. Here we use a tem size, the particle density reveals that we initially still
time step of ∆t = 2 × 10−3 which is determined to be have two clearly separated wave packets which propagate
small enough to avoid noticeable numerical errors for the towards each other during the evolution and eventually
time scales we simulate. This allows us to easily reach pass through one other. Inspection of the site-resolved
a system size of 20 qubits, the results for which for the entropy, ∆S1 (n, t), in Fig. 5(b), there is no entanglement
site resolved particle density throughout the evolution for between the wave packets at the beginning of the evo-
various masses are shown in Fig. 4. The particle density lution but as they approach one another, they become
also become entangled. However, looking at the sub-
(a) (b) (c) tracted von Neumann entropy, ∆S2 (n, t), in Fig. 5(b)
m: 0.5, g: 0.0 m: 0.8, g: 0.0 m: 1.0, g: 0.0 which corresponds to subtracting the sum of the entropies
80 80 80
obtained from evolving the fermion and the antifermion
0.4
wave packet individually, we see that this quantity is close
60 60 60 to zero throughout the entire evolution. This demon-
0.2 strates that there is essentially no excess entropy pro-
duced compared to evolving each wave packet individu-
∆hξn† ξnit
time t
−0.2
20 20 20 B. Nonvanishing coupling: the interacting case
−0.4
0 0 0 Our approach can also be applied to the interacting
0 5 10 15 0 5 10 15 0 5 10 15 case, and here we study the dynamics of the lattice
n n n Thirring model with nonvanishing coupling, g ̸= 0. To
show that the operators in Eq. (22) indeed create a wave
FIG. 4. Site-resolved particle density as a function of time packet corresponding to a particle-antiparticle pair to
for the noninteracting case g = 0 and number of lattice sites a good approximation, we first perform a direct evolu-
N = 20. The different panels correspond to different masses
tion using a Taylor expansion up to second order as for
m = 0.5 (a), 0.8 (b), and 1.0 (c). For all simulations, the ini-
tial Gaussian fermion wave packet is located around µcn = 4
the noninteracting case. The results for a system with
with mean momentum µck = 2 × 2π/N , and the initial Gaus- N = 20 sites and time step ∆t = 1 × 10−3 for various
sian antifermion wave packet is located around µdn = 15 with values of the coupling are shown in Fig. 6. Again, we
mean momentum µdk = −2 × 2π/N . Both the fermion and have again checked that the step size is small enough to
antifermion wave packets have a width of σk = 2π/N in mo- prevent any significant numerical errors.
mentum space. For t = 0 we clearly see two wave packets correspond-
ing to a particle and an antiparticle, again appearing
clearly shows the presence of the two wave packets where respectively as an excess and a deficiency in particle den-
the fermion (antifermion) manifests itself in an excess sity compared to the ground state; c.f. Fig. 6(a). More-
(lack) of particle density compared to the ground state. over, as time progresses, we observe the wave packets
Moreover, we see the desired effect that the wave pack- are moving towards each other before eventually collid-
ets are moving towards each other, where the dynamics ing and interacting. For small absolute values of the cou-
9
20 20 20 0.6
0.2 1.5
∆S1(n, t)
∆S2(n, t)
∆hξn† ξnit
time t
0.0 0.4
1.0
10 −0.2 10 10
0.2
0.5
−0.4
FIG. 5. Particle density (a), entropy ∆S1 (n, t) (b) and subtracted entropy ∆S2 (n, t) (c) for the non-interacting system with
14 sites and m = 0.8. The initial Gaussian fermion (antifermion) wave packet is located around the site n = 2 (n = 11) with
the mean momentum µck = 2π/N (µdk = −2π/N ). Both wave packets have the same width in momentum space σk = 2π/N .
0.2
∆hξn† ξnit
time t
20 20 20 20 0.0
−0.2
−0.4
0 0 0 0
0 5 10 15 0 5 10 15 0 5 10 15 0 5 10 15
n n n n
FIG. 6. Site resolved particle density as a function of time for m = 1.0 and N = 20. The different panels correspond to different
couplings g = −0.8 (a), −0.2 (b), 0.2 (c), and 0.8 (d). For all simulations, the initial Gaussian fermion wave packet is located
around µcn = 4 with mean momentum µck = 2 × 2π/N , and the initial Gaussian antifermion wave packet is located around
µdn = 15 with mean momentum µdk = −2 × 2π/N . Both the fermion and antifermion wave packets have a width of σk = 2π/N
in momentum space.
pling g, we observe a similar behavior as in the free case: havior. The wave packets are no longer passing though
the wave packets essentially pass through each other, as each other, but instead repel each other, as can be seen
shown in Figs. 6(b) and 6(c). Comparing this to the in Figs. 6(a) and 6(d).
noninteracting case (shown in Fig. 4(c)), the wave pack-
ets disperse slightly stronger for a nonvanishing coupling Our observations from the direct simulation are in
of |g| = 0.2, and the particle density spreading at t ≈ 30 agreement with the theoretical prediction for the contin-
is more pronounced. At larger absolute values of the uum model, with no observed particle production during
coupling, |g| = 0.8, we observe a noticeable change in be- the scattering process. Due to momentum conservation,
the wave packets representing the (anti)particles either
10
pass through each other or repel each other, depending the operators we derived allow for creating wave pack-
on the coupling, g. While the former behavior is ob- ets corresponding to a fermion and an antifermion with
served for couplings close to zero, the latter happens if opposite momenta even in the interacting regime. The
the absolute value of g is close to 1. (anti)particle wave packets move towards each other and
Again, we also examine the von Neumann entropy a start interacting, which is reflected in ∆S2 (n, t).
slightly smaller system size N = 14, where we use state
vector simulation in Qiskit [58] to get the vectors corre-
sponding to the four unitary summands of Eq. (24) and C. The free fermionic case on quantum hardware
get the quantum state |ψ(t)⟩ through their combination,
and we normalize the state |ψ(t)⟩ before calculating the Finally, we demonstrate that our approach is feasible
von Neumann entropy. For the time evolution operator, on current quantum hardware, and implement the nonin-
we use the matrix representation of e−iHt in our simu- teracting case on IBM’s quantum devices. To this end, we
lation and the resource estimation of Trotter decompo- use a simple parametric circuit for preparing the ground
sition is left for further work. The results for various state of the model with high fidelity, where the param-
values of the coupling are depicted in Fig. 7. Despite the eters have been optimized using a classical simulation2 .
reduced system size, the particle density in panels (a) - Subsequently, we evolve the initial state with two wave
(d) clearly shows the same effect as for N = 20: for val- packets using the techniques for the free fermionic case,
ues of the coupling close to zero the wave packets simply g = 0, discussed in Sec. III C. This allows us to simulate
pass through each other, whereas for |g| = 0.8 they re- the evolution for arbitrary times t at a constant circuit
pel each other. Focusing on the entropy in Fig. 7(e) - depth. In order to mitigate the effects of hardware noise,
7(h), we see that there is a noticeable amount of excess we employed probabilistic error amplification (PEA) [59]
entropy compared to the ground state. In particular, with zero-noise extrapolation limit [45, 60]. For PEA, we
then entropy is growing with |g|, where for positive val- applied Pauli twirling to each unique layer in our nominal
ues of the coupling, we generally observe larger values circuit. We then generated 300 circuit instances at gain
of ∆S1 (n, t) than for the corresponding negative value factors G = {1, 2, 3} and sampled each of the resulting
as a comparison between Figs. 7(e) and 7(h) as well as 900 circuits 1024 times to determine the average value of
7(f) and 7(g) reveals. While the values for the excess each observable of interest. Finally, we used an exponen-
entropy compared to the noninteracting case are larger, tial fit of the averages of the measured values at different
the evolution of ∆S1 (n, t) looks qualitatively similar to G to obtain an estimate of their zero-noise value.
the noninteracting case, as a comparison with Fig. 5(h) Figure 8 shows results for the particle density obtained
shows. on ibm peekskill for 12 qubits in comparison with the
Turning towards ∆S2 (n, t) in Figs. 7(i) - 7(l), we now exact results for several time steps. Comparing panels
observe a clear difference between the non-interacting 8(a) and 8(b), we observe good agreement between the
case and g ̸= 0. While the non-interacting case essen- data from the quantum hardware and the exact solution.
tially shows a homogeneous value of ∆S2 (n, t) through- In particular, focusing on the time slices obtained from
out the entire evolution, for g ̸= 0 we see a noticeable the quantum device in Fig. 8(b), we clearly see the two
increase of ∆S2 (n, t) at the point where the two wave wave packets moving towards each other (t = 0, 6), before
packets collide with each other. Considering the same they eventually passing through (t = 12) and start to
absolute value for the coupling, we generally observe that move away from each other again (t = 18, 24).
∆S2 (n, t) grows faster for the positive choice of g than To get a more detailed picture of the hardware results,
for the negative value. This is in agreement with obser- we show the site-resolved data for individual time slices
vation in the particle density, that for a fixed value of |g|, in Fig. 9. As the figure reveals, there is generally good
the collision happens at earlier times for a positive value agreement between the exact result and the experimental
than for the corresponding negative value (compare, e.g., data from the quantum device, in most cases both results
Figs. 7(a) and 7(d)). Interestingly, for the time scales we agree within error bar. In particular, the time-slices for
study, the data for ∆S2 (n, t) for g = 0.8 shows a qualita- the particle density illustrate once more the presence of
tively different behavior than for g = −0.8. For the posi- two separated wave packets, showing up as a peak in the
tive value, we see a peak in ∆S2 (n, t) in the center of the particle density around site 2 and a dip at site 9 for t = 0
system around t = 8 when the wave packets first start to (c.f. Fig. 9(a)). These are moving towards the center as
collide. This peak subsequently decays before two peaks time progresses resulting in a particle density noticeably
at close to the boundary of the system emerge around different from zero in the center of the system, as shown
t = 20, at which time the wave packets start to interact in Fig. 9(b). Eventually, after the two wave packets pass
again due to the periodic boundary conditions. In con-
trast, for g = −0.8 we observe a considerable increase in
∆S2 (n, t) between the wave packets after colliding, and 2 Note that finding the optimal parameters could be done with a
there is no substantial decay during the entire time we standard VQE approach. Since our primary focus in this work
simulate. lies on the time evolution, we use a classical simulation to save
In summary, our numerical results demonstrate that time on the hardware.
11
0.4
20 20 20 20 0.2
∆hξn† ξnit
time t
0.0
10 10 10 10 −0.2
−0.4
−0.6
0 0 0 0
0 5 10 0 5 10 0 5 10 0 5 10
(e) (f) (g) (h)
30 30 30 30 2.5
2.0
20 20 20 20 1.5
∆S1(n, t)
time t
1.0
10 10 10 10
0.5
0.0
0 0 0 0
0 5 10 0 5 10 0 5 10 0 5 10
(i) (j) (k) (l)
30 30 30 30
0.8
20 20 20 20 0.6
∆S2(n, t)
time t
0.4
10 10 10 10
0.2
0.0
0 0 0 0
0 5 10 0 5 10 0 5 10 0 5 10
n n n n
FIG. 7. Particle density (first row), entropy ∆S1 (n, t) (second row) and subtracted entropy ∆S2 (n, t) (third column) for
an interacting system with 14 sites and m = 0.8. The different columns correspond to different values of the coupling
g = ±0.8, ±0.2. The initial Gaussian fermion (antifermion) wave packet is located around the site n = 2 (n = 11) with
the mean momentum µck = 2π/N (µdk = −2π/N ). Both wave packets have the same width in momentum space σk = 2π/N .
through each other, we observe again a well separated V. CONCLUSION AND OUTLOOK
peak and a dip in the particle density around t = 18 in
Fig. 9(c).
In this work, we proposed a framework for studying
fermion scattering on digital quantum computing. Using
the lattice Thirring model as an example, we demon-
12
0.2 0.3
20 20
∆hξn† ξnit
∆hξn† ξnit
time t
0.0
0.0
10 10 −0.2
−0.4 -0.3
ideal simulation
0 0 hardware run
0 5 10 0 5 10
position n position n
-0.6
0 2 4 6 8 10
FIG. 8. (a) particle density for N = 12 and m = 1.0 as a 0.6
function of time obtained from an ideal simulation. (b) Same (b) t = 12
as panel (a), but with the time steps at t = 0, 6, 12, 18, 24
replaced by the ones obtained from the quantum hardware 0.3
(indicated by the gray dashed boxes).
∆hξn† ξnit
0.0
vation Foundation under the project “Quantum Com- This work is supported with funds from the Ministry of
puting for Lattice Gauge Theories (QC4LGT)”, contract Science, Research and Culture of the State of Branden-
no. EXCELLENCE/0421/0019. A. C. is supported in burg within the Centre for Quantum Technologies and
part by the Helmholtz Association —“Innopool Project Applications (CQTA).
Variational Quantum Computer Simulations (VQCS).”
This work is funded by the European Union’s Horizon
Europe Framework Programme (HORIZON) under the
ERA Chair scheme with grant agreement no. 101087126.
[1] G. A. et al., Observation of a new particle in the search tangero, Entanglement generation in (1 + 1)D qed scat-
for the standard model higgs boson with the atlas detec- tering processes, Phys. Rev. D 104, 114501 (2021).
tor at the lhc, Phys. Lett. B 716, 1 (2012). [16] T. Felser, P. Silvi, M. Collura, and S. Montangero, Two-
[2] S. C. et al., Observation of a new boson at a mass of 125 Dimensional Quantum-Link Lattice Quantum Electro-
gev with the cms experiment at the lhc, Phys. Lett. B dynamics at Finite Density, Phys. Rev. X 10, 041040
716, 30 (2012). (2020).
[3] S. Dürr, Z. Fodor, J. Frison, C. Hoelbling, R. Hoffmann, [17] G. Magnifico, T. Felser, P. Silvi, and S. Montangero, Lat-
S. D. Katz, S. Krieg, T. Kurth, L. Lellouch, T. Lippert, tice quantum electrodynamics in (3+1)-dimensions at fi-
K. K. Szabo, and G. Vulvert, Ab initio determination of nite density with tensor networks, Nat. Commun. 12,
light hadron masses, Science 322, 1224 (2008). 3600 (2021).
[4] C. Alexandrou, Hadron properties from lattice QCD, [18] P. Calabrese and J. Cardy, Evolution of entanglement
Journal of Physics: Conference Series 562, 012007 entropy in one-dimensional systems, J. Stat. Mech. 2005,
(2014). P04010 (2005).
[5] F. Hebenstreit, J. Berges, and D. Gelfand, Simulating [19] N. Schuch, M. M. Wolf, F. Verstraete, and J. I. Cirac, En-
fermion production in 1+1 dimensional qed, Phys. Rev. tropy scaling and simulability by matrix product states,
D 87, 105006 (2013). Phys. Rev. Lett. 100, 030504 (2008).
[6] F. Hebenstreit, J. Berges, and D. Gelfand, Real-time dy- [20] J. Schachenmayer, B. P. Lanyon, C. F. Roos, and A. J.
namics of string breaking, Phys. Rev. Lett. 111, 201601 Daley, Entanglement growth in quench dynamics with
(2013). variable range interactions, Phys. Rev. X 3, 031015
[7] M. C. Bañuls and K. Cichy, Review on novel methods (2013).
for lattice gauge theories, Rep. Prog. Phys. 83, 024401 [21] M. C. Bañuls, M. B. Hastings, F. Verstraete, and J. I.
(2020). Cirac, Matrix product states for dynamical simulation of
[8] M. C. Bañuls, R. Blatt, J. Catani, A. Celi, J. I. Cirac, infinite chains, Phys. Rev. Lett. 102, 240603 (2009).
M. Dalmonte, L. Fallani, K. Jansen, M. Lewenstein, [22] S. Trotzky, Y.-A. Chen, A. Flesch, I. P. McCulloch,
S. Montangero, C. A. Muschik, B. Reznik, E. Rico, U. Schollwöck, J. Eisert, and I. Bloch, Probing the relax-
L. Tagliacozzo, K. V. Acoleyen, F. Verstraete, U. J. ation towards equilibrium in an isolated strongly corre-
Wiese, M. Wingate, J. Zakrzewski, and P. Zoller, Sim- lated one-dimensional bose gas, Nat. Phys. 8, 325 (2012).
ulating lattice gauge theories within quantum technolo- [23] E. A. Martinez, C. A. Muschik, P. Schindler, D. Nigg,
gies, Eur. Phys. J. D 74, 165 (2020). A. Erhard, M. Heyl, P. Hauke, M. Dalmonte, T. Monz,
[9] M. C. Bañuls, K. Cichy, J. I. Cirac, K. Jansen, and P. Zoller, and R. Blatt, Real-time dynamics of lattice
S. Kühn, Tensor networks and their use for lattice gauge gauge theories with a few-qubit quantum computer, Na-
theories, PoS(LATTICE 2018)022 , (2019). ture 534, 516 (2016).
[10] R. Orús, A practical introduction to tensor networks: [24] N. Klco, E. F. Dumitrescu, A. J. McCaskey, T. D.
Matrix product states and projected entangled pair Morris, R. C. Pooser, M. Sanz, E. Solano, P. Lougov-
states, Ann. Phys. 349, 117 (2014). ski, and M. J. Savage, Quantum-classical computation
[11] J. C. Bridgeman and C. T. Chubb, Hand-waving and in- of schwinger model dynamics using quantum computers,
terpretive dance: an introductory course on tensor net- Phys. Rev. A 98, 032331 (2018).
works, J. Phys. A: Math. Theor. 50, 223001 (2017). [25] Y. Y. Atas, J. F. Haase, J. Zhang, V. Wei, S. M.-
[12] M. C. Bañuls, Tensor network algorithms: A route map, L. Pfaendler, R. Lewis, and C. A. Muschik, Simulating
Annual Review of Condensed Matter Physics 14, 173 one-dimensional quantum chromodynamics on a quan-
(2023). tum computer: Real-time evolutions of tetra- and pen-
[13] B. Buyens, J. Haegeman, F. Hebenstreit, F. Verstraete, taquarks, Phys. Rev. Res. 5, 033184 (2023).
and K. Van Acoleyen, Real-time simulation of the [26] W. E. Thirring, A soluble relativistic field theory, Annals
schwinger effect with matrix product states, Phys. Rev. of Physics 3, 91 (1958).
D 96, 114501 (2017). [27] D. Wecker, M. B. Hastings, N. Wiebe, B. K. Clark,
[14] S. Kühn, E. Zohar, J. Cirac, and M. C. Bañuls, Non- C. Nayak, and M. Troyer, Solving strongly correlated
abelian string breaking phenomena with matrix product electron models on a quantum computer, Phys. Rev. A
states, J. High Energy Phys. 2015 (7), 130. 92, 062318 (2015).
[15] M. Rigobello, S. Notarnicola, G. Magnifico, and S. Mon- [28] I. D. Kivlichan, J. McClean, N. Wiebe, C. Gidney,
14
A. Aspuru-Guzik, G. K.-L. Chan, and R. Babbush, dynamics for quantum systems with finite-dimensional
Quantum simulation of electronic structure with linear state spaces, Phys. Rev. A 66, 022317 (2002).
depth and connectivity, Phys. Rev. Lett. 120, 110501 [44] A. Miessen, P. J. Ollitrault, F. Tacchino, and I. Taver-
(2018). nelli, Quantum algorithms for quantum dynamics, Na-
[29] Z. Jiang, K. J. Sung, K. Kechedzhi, V. N. Smelyanskiy, ture Computational Science 3, 25 (2023).
and S. Boixo, Quantum algorithms to simulate many- [45] Y. Li and S. C. Benjamin, Efficient variational quantum
body physics of correlated fermions, Phys. Rev. Appl. 9, simulator incorporating active error minimization, Phys.
044036 (2018). Rev. X 7, 021050 (2017).
[30] A. Peruzzo, J. McClean, P. Shadbolt, M.-H. Yung, X.-Q. [46] S. Barison, F. Vicentini, and G. Carleo, An efficient quan-
Zhou, P. J. Love, A. Aspuru-Guzik, and J. L. O’Brien, tum algorithm for the time evolution of parameterized
A variational eigenvalue solver on a photonic quantum circuits, Quantum 5, 512 (2021).
processor, Nat. Commun. 5, 1 (2014). [47] A. Chiesa, F. Tacchino, M. Grossi, P. Santini, I. Tav-
[31] M. C. Bañuls, K. Cichy, Y.-J. Kao, C.-J. D. Lin, Y.-P. ernelli, D. Gerace, and S. Carretta, Quantum hardware
Lin, and D. T.-L. Tan, Phase structure of the (1 + 1)- simulating four-dimensional inelastic neutron scattering,
dimensional massive thirring model from matrix product Nature Physics 15, 455 (2019).
states, Phys. Rev. D 100, 094504 (2019). [48] S. Endo, I. Kurata, and Y. O. Nakagawa, Calculation of
[32] S. Coleman, Quantum sine-gordon equation as the mas- the green’s function on near-term quantum computers,
sive thirring model, Phys. Rev. D 11, 2088 (1975). Phys. Rev. Res. 2, 033281 (2020).
[33] J. Kogut and L. Susskind, Hamiltonian formulation of [49] A. Luther, Eigenvalue spectrum of interacting massive
wilson’s lattice gauge theories, Phys. Rev. D 11, 395 fermions in one dimension, Phys. Rev. B 14, 2153 (1976).
(1975). [50] V. E. Korepin, DIRECT CALCULATION OF THE
[34] L. Susskind, Lattice fermions, Phys. Rev. D 16, 3031 S MATRIX IN THE MASSIVE THIRRING MODEL,
(1977). Theor. Math. Phys. 41, 953 (1979).
[35] E. Lieb, T. Schultz, and D. Mattis, Two soluble models [51] V. E. Korepin, The mass spectrum and the S matrix of
of an antiferromagnetic chain, Ann. Phys. 16, 407 (1961). the massive Thirring model in the repulsive case, Com-
[36] F. Verstraete, J. I. Cirac, and J. I. Latorre, Quantum cir- munications in Mathematical Physics 76, 165 (1980).
cuits for strongly correlated quantum systems, Physical [52] A. Luther, Eigenvalue spectrum of interacting massive
Review A 79, 032316 (2009). fermions in one dimension, Phys. Rev. B 14, 2153 (1976).
[37] E. Zohar, J. I. Cirac, and B. Reznik, Quantum simula- [53] M. Luscher, Dynamical Charges in the Quantized, Renor-
tions of lattice gauge theories using ultracold atoms in malized Massive Thirring Model, Nucl. Phys. B 117, 475
optical lattices, Rep. Prog. Phys. 79, 014401 (2016). (1976).
[38] P. Jordan and E. P. Wigner, Über das paulische [54] S. Coleman, More about the massive schwinger model,
äquivalenzverbot (Springer, 1993). Ann. Phys. (N.Y.) 101, 239 (1976).
[39] R. Belyansky, S. Whitsitt, N. Mueller, A. Fahimniya, [55] S. Mandelstam, Soliton operators for the quantized sine-
E. R. Bennewitz, Z. Davoudi, and A. V. Gorshkov, gordon equation, Phys. Rev. D 11, 3026 (1975).
High-energy collision of quarks and hadrons in the [56] A. B. Zamolodchikov, Exact Two Particle s Matrix of
schwinger model: From tensor networks to circuit qed, Quantum Sine-Gordon Solitons, Pisma Zh. Eksp. Teor.
arXiv:2307.02522 (2023). Fiz. 25, 499 (1977).
[40] A. Peruzzo, J. McClean, P. Shadbolt, M.-H. Yung, X.-Q. [57] M. Karowski and H. J. Thun, Complete S Matrix of the
Zhou, P. J. Love, A. Aspuru-Guzik, and J. L. O’Brien, Massive Thirring Model, Nucl. Phys. B 130, 295 (1977).
A variational eigenvalue solver on a photonic quantum [58] Qiskit contributors, Qiskit: An open-source framework
processor, Nat. Commun. 5, 1 (2014). for quantum computing (2023).
[41] S. McArdle, T. Jones, S. Endo, Y. Li, S. C. Benjamin, [59] Y. Kim, A. Eddins, S. Anand, K. X. Wei, E. van den
and X. Yuan, Variational ansatz-based quantum simula- Berg, S. Rosenblatt, H. Nayfeh, Y. Wu, M. Zaletel,
tion of imaginary time evolution, npj Quantum Informa- K. Temme, and A. Kandala, Evidence for the utility of
tion 5, (2019). quantum computing before fault tolerance, Nature 618,
[42] J. Gacon, J. Nys, R. Rossi, S. Woerner, and G. Carleo, 500 (2023).
Variational quantum time evolution without the quan- [60] K. Temme, S. Bravyi, and J. M. Gambetta, Error mitiga-
tum geometric tensor, arXiv:2303.12839 (2023). tion for short-depth quantum circuits, Phys. Rev. Lett.
[43] M. A. Nielsen, M. J. Bremner, J. L. Dodd, A. M. Childs, 119, 180509 (2017).
and C. M. Dawson, Universal simulation of hamiltonian
In this section, we explain how to decompose a complex unitary matrix u based on Givens rotation. We then
demonstrate how to decompose the operation V (u) into local operations based on this decomposition of u.
Given a complex unitary matrix u with dimensions N × N , we can find a series unitary matrices rnl (θnl ) and pl (β⃗l )
15
sN −1 · · · s0 u = I, (A1)
(A2)
where
l ⃗
sl = rl+1
l l
(θl+1 l
) · · · rN l
−1 (θN −1 )p (βl ), l = 0, · · · N − 1, (A3)
sN −1 = pN −1 (β⃗N −1 ). (A4)
The matrix sl is serves to diagonalize the l-th column of matrix u. In this process, the diagonal matrix pl (β) ⃗
eliminates the phase of elements in column l and then Givens rotation matrix rnl (θ) eliminates the element unl . The
formula of matrix pl (⃗(β)) and rnl (θ) are as following:
iβ
e 0 0 ··· 0
0 eiβ2 · · · 0
⃗ =
pl (β) .. . ..
(A5)
. .. .
0 · · · · · · eiβN −1 N ×N
1 ··· 0 0 ··· 0
. . . .. ..
.. . . .. . .
l
0 · · · cos(θ) − sin(θ) · · · 0
rn (θ) =
, (A6)
0 · · · sin(θ) cos(θ) · · · 0
.. .. .. . . ..
. . . . .
0 ··· 0 0 · · · 1 N ×N
where the matrix rnl is diagonal except for (n − 1)-th, n-th row and columns.
Below, we illustrate the procedure for the diagonalization of the first column of u. Considering a general complex
unitary matrix u:
where anl , bnl are the absolute value and phase of matrix element unl . In order to diagonalize the first column of u.
We will multiply the phase matrix p0 (β) ⃗ to u such that the elements in first column of u become real. The parameters
0 ⃗ ⃗ ⃗ · u is as follows:
in p (β) are denoted as β = (b0,0 , · · · , bN −1,0 ), and the matrix p0 (β)
a0,0 a0,1 e−ib0,1 ··· a0,N −1 e−ib0,N −1
e e
. .. ..
.
⃗ ·u= .
p0 (β) . ··· .
. (A8)
aN −2,0 aN −2,1 e−ibN −2,1 · · · aN −2,N −1 e−ibN −2,N −1
e e
0 0 0 ⃗
rN −1 (θN −1 ) · p (β) · u
a0,0 a0,1 e−ib0,1 ··· a0,N −1 e−ib0,N −1
e e
.. .. .. ..
=
. . . .
0 0 −ibN −2,1 −ibN −2,N −1
cos θN −1 aN −2,0 − sin θN −1 aN −1,0 e aN −2,1 e ··· e aN −2,N −1 e
e e
0 0 −iebN −1,1 −ie
bN −1,N −1
cos θN −1 aN −1,0 + sin θN −1 aN −2,0 e aN −1,1 e ··· e aN −1,N −1 e N ×N (A9)
−ie
b0,1 −ie
b0,N −1
a0,0 a0,1 e ··· a0,N −1 e
... .. ..
a
..
0 N −1,0
θN −1 =arctan − aN −2,0
−→ . . .
aN −2,1 e−ibN −2,1 · · · e
aN −2,0 e aN −2,N −1 e−ibN −2,N −1
e e
e
0 aN −1,1 e−ibN −1,1 · · · e aN −1,N −1 e−ibN −1,N −1 N ×N
e e
e
In the resulting matrix above, all the matrix elements in (N − 1)-th row and (N − 2)-th row are changed due to the
Givens rotation. To simplify, we use e anl and ebnl to represent the absolute value and phase of matrix elements after
the rotation. To diagonalize the first column of the matrix u, we need to multiply more Givens rotation matrix to u
to zero the matrix elements on the row N − 2, N − 3, · · · , 1 in first column:
0 ⃗
s0 · u = r10 (θ10 ) · · · rN
0 0
−1 (θN −1 ) · p (β) · u
1 0 ··· 0
0 a1,1 e−ib1,1 ··· a1,N −1 e−ib1,N −1
e e
a
e e
0
θn =arctan − a n,0 (A10)
n−1,0
. .. . ..
. .
−→ .
. . .
,
0 e aN −2,1 e−ibN −2,1 · · · e aN −2,N −1 e−ibN −2,N −1
e e
an,0
where θn0 = arctan − an−1,0 for n = N − 1, N − 2, · · · , 1. Furthermore, the first elements of other columns are also
zeros after diagonalizing the first column. This occurs due to the orthogonality between different columns in a unitary
matrix. The procedure is similar if we want to diagonal the second column of u, and so on... In the wave packet
c(d)
preparation, we specifically focus on the first column of u, denoted by un0 = ϕn . Therefore, assuming s0 · u = I,
the decomposition of u will be:
† † 0 0 †
u = (s0 )† = p0 (β) ⃗ 0 0
· rN −1 (θN −1 ) · · · r1 (θ1 ) . (A11)
Regarding the unitary operator V (u) defined in Eq. (16), one can get a decomposition of it based on the property
V (u · u′ ) = V (u) · V (u′ ) and the decomposition of matrix u:
with,
! !
X X
V † (p0 ) = exp − [log p0 ]nl ξn† ξl = exp −i βn ξn† ξn ,
nl n
! (A13)
[log rn0 ]ij ξi† ξj †
X
†
V (rn0 ) = exp − = exp θn0 [ξn−1 ξn − ξn† ξn−1 ] .
nl
The final equation above employs the following result, which is non-zero only for the elements with indices (i, j) =
(n − 1, n) and (n, n − 1):
0 ··· 0 0 ··· 0
. .
.. . . ... ... ..
.
l
0 · · · 0 −θ · · · 0
log [rn (θ)] =
0 · · · θ 0
(A14)
· · · 0
.. .. .. . . ..
. . . . .
0 · · · 0 0 · · · 0 N ×N
17
In this appendix, we analytically determine the particle density noninteracting case during the scattering process
via calculations in momentum and Fock space. Since we are dealing with a free fermionic system, these computations
involve a matrices whose dimensions N × N , enabling us to numerically simulate the phenomenon for large system
and to cross-check the results in position space.
In momentum space, the free fermion Hamiltonian is diagonal in terms of the fermionic and antifermionic creation
(annihilation) operators c†k , d†k (ck , dk [15]:
2π N N
wk c†k ck − dk d†k ,
X
H= k ∈ Λk = − ,··· −1 , (B1)
L 4 4
k
and the vacuum |Ω⟩ in this theory is the state with no excitations of fermions and antifermions:
As there is no interaction between fermions and antifermions in momentum space, the wave function of the system is
as the tensor product of the wave function for fermion and antifermion subsystems:
When considering the Fock basis for the momentum state, the Gaussian wave packets for the fermion and antifermion
can be represented as the vectors
T
|ψc (t = 0)⟩ → ϕck1 , . . . , ϕckN/2 , (B4)
T
|ψd (t = 0)⟩ → ϕdk1 , . . . , ϕdkN/2 , (B5)
with ki = 2π/N ×(i − N/4 − 1). In this basis, the fermion and antifermion part of Hamiltonian correspond to diagonal
matrices with dimension N/2 and the full Hamiltonian will be a direct sum of the matrix for the fermions and the
antifermions. The fermionic part will be given by
w0 0 · · · 0
0 w2 · · · 0
wk c†k ck → .. .. . .
X
Hc = .. (B6)
k
. . . .
0 0 · · · wN/2−1
and analogously the antifermionic part. As a result, the time evolution of the initial fermionic Gaussian wave packet
can be expressed as follows (up to a global phase)
To obtain the particle density, ⟨ξn† ξn ⟩t at each spatial site, we need to transform the operator ξn† ξn into momentum
space. For even n we find
s s s s
1 X m + wp m + wq −i(p−q)n 1 X m + wp m + wq −i(p−q)n
ξn† ξn = e × c†p cq − e × d†p dq × vp vq
N wp wq N wp wq
pq∈Λk pq∈Λk
1 X m + wp
+ × vp2 , (B9)
N p wp
X
c†p × Cpq
n0
× cq + d†p × Dpq
n0
→ × dq + Ceven ,
pq∈Λk
18
where the constant Ceven and Codd are the particle density of the vacuum for even and odd sites. The definition of
wk and vk can be found in Eq. (10). Using the matrix representation in one-particle Fock space C nr , Dnr , r ∈ {0, 1},
T T
we can evaluate the particle density quickly by calculating ϕ⃗ck × C nr × ϕ⃗ck + ϕ⃗dk × Dnr × ϕ⃗dk . We show the result
with 200 sites in Fig. 10.
0.04
700
600
0.02
500
∆hξn† ξnit
time t
400 0.00
300
−0.02
200
100
−0.04
0
0 50 100 150
position n
FIG. 10. The particle density for free staggered fermion with N = 200. The initial fermionic Gaussian wave packet is located
at µcn = 30 with mean momentum µck = 20 × 2π/N ; the initial antifermion wave packet is located at N − 1 − µcn with the same
absoulte value for the momentum as the fermion but in the opposite direction. For both the fermion and antifermion wave
packet we choose a width of σk = 2π/N in momentum space.
In the noninteracting case, corresponding to g = 0 in Eq. (2), the Hamiltonian is quadratic in the fermionic operators
and thus corresponds to a free staggered Dirac fermions. The time evolution operator operator for this case reads as
where M is an antihermitian matrix in Fock space depending only on the mass and evolution time
As Eq. (C1) reveals, the operator U (t) has the same form as Eq. (16) with u = eM . Thus, we can use Givens
rotation to decompose the time evolution operator in the same way as V (u). Appendix A provides a comprehensive
explanation of how to use Givens rotation to diagonalize the first column of the unitary matrix u. The remaining
columns can then successively brought to diagonal form by repeating the procedure and eventually resulting in the
decomposition of U (t). Parallelizing the Givens rotations on distinct qubit-pairs, the time evolution operator in the
noninteracting case can be decomposed into a quantum circuit of constant depth O(4N ), using O(N 2 ) CNOT gates.